5/10/2004 - BAE Systems plc and Celoxica Ltd. announced plans for the first public demonstration of safety and security applications, in the Celoxica booth at the 41st Design Automation Conference in San Diego from June 7-11.
The demonstrations, built with BAE Systems' LEARRNN technology, Celoxica's System FPGA design tools and RC series development platforms, showcase the application of LEARRNN technology for the emerging challenges of video surveillance and biometrics. Using Celoxica's RC200 digital image and signal processing reconfigurable development platform, both applications embody library elements for users to implement video image processing and pattern-matching functions frequently required in the two safety and security applications. Customer solutions are delivered by the BAE Systems and Celoxica partnership via their integrated design tools and Celoxica professional consulting services.
The partnership's 'One to Many' fingerprint matching engine provides a match result (identification) of a test fingerprint against an enrolled database of up to 800 in less than 50ms, using the standard CFMEF fingerprint format and by fusing correlation and minutia based matching techniques. A single matching node within the demonstrator can hold many multiples of enrolled fingerprints locally, and its design is deliberately scalable to accommodate very large fingerprint databases by adding further matching nodes in parallel.
"LEARRNN is ideally suited to accelerating biometric identification applications because it provides an economic, pure logic correlation architecture," said Bob Flint, Director, Ventures, BAE Systems. "Coupled with reconfigurable processing technology and system level design techniques, LEARRNN enables a massive number of matching processes to run in parallel with easy data fusion of the results. This performance and scalability in a compact and portable format carries many advantages and overcomes some of the current issues associated with automatic fingerprint matching and analysis."
Using the LEARRNN technique all functions are built using a combination of pure logic and memory and with custom design enabled via the DK Design Suite, product-engineered implementations in silicon are straightforward.
"With the global focus on authenticating people's identity, biometrics is an important application area. It requires an order of magnitude leap in performance if the goals of safe border control and secure access to valuable resources and assets are to be realized," commented Tony Vitucci, VP of consulting for Celoxica. "By combining our expertise, BAE Systems and Celoxica have developed a solution that is applicable to many classes of pattern matching problems."
The second demonstrator shows noise removal on real-time video using a 3x3 neural filter operator. The filter is capable of operation at more than 100 MHz pixel clock frequency on a 752 x 582 (CIF+) pixel input digitized video stream. This means it can deal with real time frame rates of 240 Hz at this resolution, or with a 1280 x 960 (e.g. HDTV) image at 60 Hz frame rate. A single filter provides a factor of 10 reduction in impulsive noise, and uses 150 LUT slices (30k gate equivalents). The filter is easily cascaded or scaled to accommodate larger or more complex operators.
The real time video filter demonstration showcases another key application area for BAE Systems and Celoxica in real-time signal and video image processing, both in security and other markets demanding high performance solutions such as broadcast, wireless mobile and automotive safety.
"Fast image processing without the need for complex math and extended design times is remarkable," Vitucci stated. "We help customers implement object detection and tracking functions with a low development overhead compared to traditional programming methods and silicon architectures. We're making the difference by providing an incredibly fast design path to implementation of pattern recognition, fault/difference detection and diagnosis/interpretation in reconfigurable hardware."
All basic LEARRNN functions and building blocks are available as a toolbox for MATLAB. Custom IP is designed using the DK Design Suite and full system co-simulation is made easy by using Celoxica's co-simulation technology for multiple concurrent simulations.
About BAE SYSTEMS
BAE SYSTEMS is an international company engaged in the development, delivery and support of advanced defence and aerospace systems in the air, on land, at sea and in space. The company designs, manufactures and supports military aircraft, surface ships, submarines, radar, avionics, communications, electronics and guided weapon systems. It is a pioneer in technology with a heritage stretching back hundreds of years. It is at the forefront of innovation, working to develop the next generation of intelligent defence systems. BAE SYSTEMS has major operations across five continents and customers in some 130 countries. The company has more than 90,000 people and generates annual sales of approximately £12 billion through its wholly owned and joint venture operations. BAE SYSTEMS, Innovating for a safer world.
An innovator in system-level electronic design automation (EDA), Celoxica supplies the design technology, IP and services that define Software-Compiled System Design, a methodology that exploits higher levels of design abstraction to dramatically improve silicon design productivity. Celoxica's products address hardware/software partitioning, co-verification and C-based synthesis to reconfigurable hardware. Established in 1996, Celoxica offers a proven route from complex software algorithms to hardware, and provides an ideal design environment for System FPGA with significant productivity advantages for digital signal processing applications such as imaging, electronic security and communications. For more information, visit: www.celoxica.com.
Celoxica, Handel-C and the Celoxica logo are trademarks of Celoxica, Ltd. LEARRNN is a trademark of BAE Systems plc.
Previous Page | News by Category | News Search
If you found this page useful, bookmark and share it on: